Add support for rasterizing geopandas dataframes directly

[philippjfr]

Should consider whether to move the geopandas interface from geoviews to holoviews, but here's it working:

import geopandas as gpd
import geoviews
import holoviews as hv

from geodatasets import get_path
from holoviews.operation.datashader import rasterize

hv.extension('bokeh')

nybb = gpd.read_file(get_path("nybb"))

polys = hv.Polygons(nybb, datatype=['geodataframe'])

rasterize(polys)

Fixes #5951

[jbednar]

Nice! My understanding from @ianthomas23 was that the geopandas rendering was slower than the spatialpandas rendering. Are there any implications that we need to deal with? E.g. do we need to let users decide between directly rendering as geoparquet and converting to spatialpandas first? If the rendering is a lot faster, converting to sp might take more time initially and use more memory but then give faster zooms. I haven't tried the two alternatives, though, so this is just speculation.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 88.39%. Comparing base (023ad71) to head (cb5949d).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #5958   +/-   ##
=======================================
  Coverage   88.39%   88.39%           
=======================================
  Files         323      323           
  Lines       67609    67610    +1     
=======================================
+ Hits        59761    59762    +1     
  Misses       7848     7848           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ianthomas23]

We are in a difficult situation here with regard to automatic conversion from geopandas to spatialpandas or not. If we had never previously converted then we would be fine to always leave the data in the format passed to us, and provide documentation on how to convert between the two. But if we turn off automatic conversion from geopandas to spatialpandas then users with a certain class of problem (data above a certain size) will experience a performance loss.

[jbednar]

Yes, but we can also go by whether SpatialPandas is installed, using it if it's installed and passing through to GeoPandas support if it isn't. I think that should cover most cases right now, and if eventually we stopped using SpatialPandas, that branch would never get executed, and we can silently move on.

[philippjfr]

The explicit thing to do is to respect what the user provided, if they passed in geopandas we should keep in geopandas.